better contextual suggestions by applying domain knowledge

Better Contextual Suggestions by Applying Domain Knowledge

Thaer Samar, Alejandro Bellogin, Arjen P. de Vries

Contextual Suggestions

Given a user profile and a context, make suggestions AKA Context-aware Recommendation, zero-query

Information Retrieval, …

“Entertain me”

Recommend “things to do”, where User profile consists of opinions about attractions

Context consists of a specific geo-location

My Profile

My Context

My Suggestions

WORM

Poortgebouw

TREC Contextual Suggestions (1/3)

Given a user profile 70 – 100 POIs represented by a title, description

and URL (situated in Chicago / Santa Fe)

Rated on a scale 0 – 4

125, Adler Planetarium & Astronomy Museum, ''Interactive exhibits & high-tech sky shows entertain stargazers -- lakefront views are a bonus.'',

http://www.adlerplanetarium.org/131,Lincoln Park Zoo,"Lincoln Park Zoo is a free 35-acre zoo

located in Lincoln Park in Chicago, Illinois. The zoo was founded in 1868, making it one of the oldest zoos in the U.S. It is also one of a few free admission zoos in the United States.", http://www.lpzoo.org/

700, 125, 4, 4700, 131, 0, 1

TREC Contextual Suggestions (2/3)

… and a context Corresponding to a metropolitan area in the USA,

e.g., 109, Kalamazoo, MI

TREC Contextual Suggestions (3/3)

Suggest Web pages / snippets From the Open Web, or from ClueWeb

700, 109 ,1,"About KIA History Kalamazoo Institute of Arts KIA History","The Kalamazoo Institute of Arts is a nonprofit art museum and school. Since , the institute has offered art classes and free admission programming, including exhibitions, lectures, events, activities and a permanent collection. The KIAs mission is to cultivate the creation and appreciation of the visual arts for the communities",clueweb12-1811wb-14-09165

Approach

For a given location, select candidate web pages from Clueweb

Rank the candidates based on their cosine-similarity to the POIs in the user profile (separated in a positive and a negative profile)

Snippet Generation

Generate POI title: Extract <title> or <header> tags

Generate personalized POI description: Extract <description> tag Break documents into sentences, ranked on their

similarity with the user profile Concatenate until 512 bytes reached

Candidate selection

In 2013, the CWI Clueweb based run ranked far below all other (Open Web) runs A few issues related to evaluation, see our ECIR

2014 short paper

But, also, the commercial Open Web search engines (Google, Bing or Yahoo!) return much better candidates for queries derived from the context than we did

Geo-Filtering

Exact mention of given context Format: {City, ST} e.g., Miami, FL

Exclude documents that mention multiple contexts E.g., a Wikipedia page about cities in Florida state

Domain Knowledge (1/2)

Point-of-Interest heuristic: POIs will be represented on the major tourist

information sites

{yelp, tripadvisor, wikitravel, zagat, xpedia, orbitz, and travel.yahoo}

Extract the Clueweb documents from these domains (TouristListFiltered) E.g., http://www.zagat.com/miami

Expand with the outlinks also contained in ClueWeb12 (TouristOutlinksFiltered)

Domain Knowledge (2/2)

Use Foursquare API to identify the URLs of POIs for the given context

If the POI has no corresponding URL, use Google API with a query using foursquare POI + context, i.e., “Cortés Restaurant Miami, FL”

Extract any document from Clueweb whose host matches the (1,454 unique) hosts of the URLs identified (AttractionFiltered)

50 attractions per contextFormat: attraction name, URL e.g., Cortés Restaurant, http://cortesrestaurant.com

Miami, FL

Candidate Selection

ClueWeb12

733,019,372 web pages

“City, ST”8,883,068 docs

TouristListFiltered (175,260)

TouristOutlinksFiltered (97,678)

AttractionsFiltered (102,604)

GeoFiltered

TouristFiltered

Overall Results

Note: P@5 and MRR consider three dimensions of relevance:

geographical (geo), description (desc) and document (doc) relevance

TouristFiltered >>

GeoFiltered

(I.e., TouristFiltered suggests better POIs for 33.1% of the

judged topics)

TouristFiltered vs. GeoFiltered

% topics

Decompose metrics

Ignoring geo-relevance:

GeoFiltered ~ TouristFiltered

Decompose metrics

Ignore geo-relevance:

Geo-relevance only:

The two runs have almost similar

performance in the desc and doc dimensions

TouristFiltered is more

geographically appropriate

Type of domain knowledge

TouristFiltered consists of three parts: TouristListFiltered (TLF)

TouristOutlinksFiltered (TOF)

AttractionFiltered (AF)

Foursquare gives the most significant improvement in

performance

Conclusions

Domain knowledge about sites that are more likely to offer attractions lead to better suggestions

The best results were obtained when identifying attractions through specialized services such as Foursquare

Next Steps

Improve our recommendation algorithm E.g., weighted candidate selection

Understand the remaining difference with Open Web based results Our Clueweb results are reproduceable but not

yet as good